Limited "generics" in C without macros or UB June 5, 2017 on Drew DeVault's blog

I should start this post off by clarifying that what I have to show you today is not, in fact, generics. However, it’s useful in some situations to solve the same problems that generics might. This is a pattern I’ve started using to reduce the number of void* pointers floating around in my code: multiple definitions of a struct.

Errata: we rolled this approach back in wlroots because it causes problems with LTO. I no longer recommend it.

Let’s take a look at a specific example. In wlroots, wlr_output is a generic type that can be implemented by any number of backends, like DRM (direct rendering manager), wayland windows, X11 windows, RDP outputs, etc. The wlr/types.h header includes this structure:

struct wlr_output_impl;
struct wlr_output_state;

struct wlr_output {
    const struct wlr_output_impl *impl;
    struct wlr_output_state *state;
    // [...]

void wlr_output_enable(struct wlr_output *output, bool enable);
bool wlr_output_set_mode(struct wlr_output *output,
    struct wlr_output_mode *mode);
void wlr_output_destroy(struct wlr_output *output);

wlr_output_impl is defined elsewhere:

struct wlr_output_impl {
    void (*enable)(struct wlr_output_state *state, bool enable);
    bool (*set_mode)(struct wlr_output_state *state,
        struct wlr_output_mode *mode);
    void (*destroy)(struct wlr_output_state *state);

struct wlr_output *wlr_output_create(struct wlr_output_impl *impl,
        struct wlr_output_state *state);
void wlr_output_free(struct wlr_output *output);

Nowhere, however, is wlr_output_state defined. It’s left an incomplete type throughout all of the common wlr_output code. The “generic” part is that each output implementation, in its own private headers, defines the wlr_output_state struct for itself, like the DRM backend:

struct wlr_output_state {
    uint32_t connector;
    char name[16];
    uint32_t crtc;
    drmModeCrtc *old_crtc;
    struct wlr_drm_renderer *renderer;
    struct gbm_surface *gbm;
    EGLSurface *egl;
    bool pageflip_pending;
    enum wlr_drm_output_state state;
    // [...]

This allows implementations of the enable, set_mode, and destroy functions to avoid casting a void* to the appropriate type:

static struct wlr_output_impl output_impl = {
    .enable = wlr_drm_output_enable,
    // [...]

static void wlr_drm_output_enable(struct wlr_output_state *output,
        bool enable) {
    struct wlr_backend_state *state =
        wl_container_of(output->renderer, state, renderer);
    if (output->state != DRM_OUTPUT_CONNECTED) {
    if (enable) {
        // [...]
    } else {

// [...]
struct wlr_output output = wlr_output_create(&output_impl, output);

The limitations of this approach are apparent: you cannot work with multiple definitions of wlr_output_state in the same file. However, you get improved type safety, have to write less code, and improve readability.

Have a comment on one of my posts? Start a discussion in my public inbox by sending an email to ~sircmpwn/ [mailing list etiquette]

Articles from blogs I read Generated by openring

Saying "no" to unethical tasks

Back in spring 2019, I was a student working as an intern at the Intel Open Source Graphics Center in Finland. I was mainly focused on improving igt-gpu-tools, the test suite that runs each time a patch is submitted for the i915 kernel driver. I really liked…

via emersion November 23, 2020

SourceHut's second year in alpha

Today is the second anniversary of SourceHut being made available to the general public during its alpha phase of development, and the end of the fourth year of development. What a year it’s been! If only I could send a warning to the younger me who wrote la…

via Blogs on Sourcehut November 15, 2020

Graphics for JVM

Introducing Skija, Java bindings for Google graphics library, Skia

via November 14, 2020